The Additional sex combs gene of Drosophila encodes a chromatin protein that binds to shared and unique Polycomb group sites on polytene chromosomes

Early in development, the homeotic genes of Drosophila are regulated by transiently expressed transcription factors encoded by the segmentation genes (White and Lehmann, 1986; Harding and Levine, 1988; Irish et al., 1989). Subsequently, the Polycomb group (PcG) genes are required to stably maintain repression of homeotic loci in cells where the target genes were initially repressed (Struhl and Akam, 1985; Wedeen et al., 1986; Jones and Gelbart, 1993; McKeon and Brock, 1991; Simon et al., 1992). The mechanism of PcG repression is not understood at the molecular level. Based on the similarity of a domain found in POLYCOMB and HP1, encoded by the Su(var)205 locus, it was proposed that PcG proteins might regulate chromatin structure to make it more heterochromatin-like (Paro, 1990; Alberts and Sternglanz, 1990). However, other models have been proposed: (1) a compartmentalization model (Schlossherr et al., 1994; Strouboulis and Wolffe, 1996), (2) a looping model in which interactions between PcG proteins cause looping out of DNA that interferes with interactions between enhancers and promoters (Pirrotta, 1995), and (3) a repressor model in which PcG proteins interfere with basal transcription (Bienz, 1992).

About 15 PcG genes have been characterized (see Simon, 1995 for a review), although it has been estimated that there may be as many as 40 PcG genes (Jurgens, 1985; Landecker et al., 1994). Mutations in one PcG gene usually enhance the phenotypes of mutations in other PcG genes (Jurgens, 1985; Cheng et al., 1994; Campbell et al., 1995), suggesting that PcG proteins have similar functions, or are members of a complex (Locke et al., 1988; see Kennison, 1995 for a review). The cloned PcG genes encode chromatin proteins that bind chromosomes in a DNA-dependent manner. The distributions of Polycomb (PC), polyhomeotic (PH) and Polycomblike (PCL) protein binding completely overlap at about 100 sites on polytene chromosomes (Zink and Paro, 1989; DeCamillis et al., 1992; Lonie et al., 1994), and the distributions of Posterior sex combs (PSC) and Enhancer of zeste overlap at a subset of these sites (Rastelli et al., 1993; Carrington and Jones, 1996).In addition PC and PH coimmunoprecipitate, and PC and PH bind PSC in a yeast two-hybrid assay and in vitro (M. K and H. W. B., unpublished data) providing molecular support for the suggestion that PcG proteins form a multimeric complex (Franke et al., 1992). Nevertheless, there must be discrete PcG complexes, because the distributions of PcG proteins on polytene chromosomes are not identical. Mammalian homologues of PcG genes have been cloned, and these proteins also form complexes and are required for regulation of homeotic genes (Alkema et al., 1995, 1997; Core et al., 1997; Akasaka et al., 1996).

Additional sex combs (Asx) shares many properties with other members of the PcG. It was first described by Jurgens (1985). Mutant alleles were recovered because males had a dominant extra sex combs phenotype and because of head defects in embryos. Embryos that are homozygous mutant for Asx exhibit relatively mild posterior transformations of thoracic and abdominal segments, and partial failure of head involution so that the cephalopharyngeal apparatus is more anterior and mouth hooks are more lateral than in wild-type embryos. Even though most segments are transformed posteriorly, the extent of the transformations is less severe than those seen in Pc⁻ embryos (Breen and Duncan, 1986). In mutant embryos derived from homozygous mutant mothers, the posterior transformations are more extensive. All the thoracic denticle belts partially resemble abdominal denticle belts, Keilin’s organs are reduced or absent and the abdominal segments are transformed posteriorly. Nevertheless, even in embryos derived from maternal mutants, the extent of posterior transformation is not complete (Breen and Duncan, 1986; Soto et al., 1995). Like other dominant PcG mutations, mutations in Asx enhance the phenotypes of other dominant PcG mutations in doubly heterozygous adults (Campbell et al., 1995) and in embryos (Jurgens, 1985). Most Asx mutations appear to be gain of function, because their zygotic phenotypes are more severe than those of homozygous deficiency embryos (Sinclair et al., 1992).

Asx mutations also exhibit a number of phenotypes that are exhibited by a restricted subset of PcG genes, or that are unique to Asx. Even though zygotic Asx mutants ectopically express homeotic genes (McKeon and Brock, 1991; Simon et al., 1992), zygotic mutant embryos and embryos derived from mothers with Asx⁻ germlines exhibit tissue-specific derepression. SCR, UBX, ABD-A and ABD-B mis-expression occurs in epidermal and visceral mesoderm cells, but generally are not mis-expressed in the central nervous system (CNS) (Soto et al., 1995). For the PcG genes that have been studied, only Pcl mutants also exhibit tissue-specific misexpression of homeotic genes (Soto et al., 1995). These authors have speculated that ASX might act on different targets than other PcG proteins. Heterozygous Asx mutant adults, and homozygous mutant embryos exhibit segment defects that affect even-numbered segments more often than odd-numbered segments (Sinclair et al., 1992). Segmentation defects have also been reported in other PcG genes, including super sex combs (sxc) (Ingham, 1984), and l(4)29/pleiohomeotic (Breen and Duncan, 1986). Together, the evidence indicates that Asx has a specific role in gene regulation.

Two other features suggest that Asx differs from most PcG genes. First, the P-element-induced allele, Asx^P1, exhibits anterior as well as posterior transformations. In addition to posterior transformation of the fourth abdominal segment towards the fifth, homozygous flies show swollen halteres with occasional bristles similar to those seen in the triple row margin of wings, and the fifth abdominal segment shows unpigmented patches, consistent with transformation towards a more anterior tergite. Anterior transformations are characteristic of mutations in the trithorax Group (trxG), genes required for the continued activation of homeotic genes (Kennison, 1995). Their presence in an Asx mutant suggests that Asx might have a dual role in repression and in activation. Consistent with this idea, mutations in Asx act as enhancers of position-effect variegation, whereas most PcG mutations do not affect position-effect variegation (Sinclair et al., 1998). It might be expected that PcG mutations would suppress position-effect variegation if they function directly to regulate chromatin structure at target loci to repress gene activity. Because Asx mutations enhance position-effect variegation, similar to at least one member of the trxG (Farkas et al., 1994), it can be argued that ASX is required to maintain an open conformation of chromatin at variegating loci.

The features reported above make Asx a particularly interesting member of the PcG, as its analysis may shed light on the role of PcG proteins in segmentation, tissue-specificity, and its role in activation versus repression of target loci. To better understand the molecular basis for ASX function, we cloned the Asx locus. It encodes a putative 1668 amino acid protein with two domains that are conserved in mammals. The RNA and protein are ubiquitous, except that Asx mRNA does not appear to be deposited in oocytes. Asx is required for proper expression of Ubx in the central nervous system. The distribution of ASX on polytene chromosomes partially overlaps that of PC, PH and PCL, but there are many binding sites unique to ASX. These data rule out the possibility that tissue-specific regulation by Asx is achieved by tissue-specific distribution of the ASX protein, but support the hypothesis that unique Asx phenotypes arise at least partly as a consequence of a target specificity different from other PcG proteins.

The Asx^P1 strain and other Asx alleles were described in Sinclair et al. (1992). All flies were raised at 25°C on standard cornmeal sucrose medium containing tegosept as a mould inhibitor. Embryos were collected from cages on agar plates spread with yeast paste to obtain staged embryos for preparation of RNA, and from egg-laying chambers for in situ hybridization and antibody staining. Transformed lines containing the bxd14 element were obtained from W. Bender.

General molecular procedures were carried out as described (Sambrook et al., 1989). The Asx^P1 mutation is a P element insertion that fails to complement Asx. DNA flanking the insertion was recovered from a genomic library established from Asx^P1 DNA using P element DNA as a probe, and used to screen a cosmid library supplied by J. Tamkun to recover cosmids spanning the insertion point. Reverse northern analysis was used to identify transcription units (Brunk et al., 1991). An imaginal disc library supplied by G. Rubin was screened to recover cDNAs corresponding to the 2 transcription units flanking the P element insertion site. Dideoxy sequencing was carried out manually on both strands of selected cDNAs, using protocols recommended by the suppliers of the DNA polymerases (Amersham, Pharmacia), and on the genomic DNA to map introns, using a combination of directed deletions and primer walking.

RNA was isolated from staged embryos using the Trizol^TM reagent (Bethesda Research Laboratories), and poly(A)⁺ RNA was isolated by two rounds of chromatography on oligo(dT) cellulose. About 2 μg of RNA for each developmental stage was fractionated in 1.0% agarose gels containing formaldehyde, transferred to nylon filters, and probed with cDNA or with RP-49 genomic DNA to control for loading.

Analysis of Asx distribution in ovaries and embryos was carried out as described (DeCamillis and Brock, 1994).

A carboxy terminal antibody was raised in rabbits, and in mice, using an Asx C-terminal fragment containing amino acid residues 1590 to 1668, subcloned as a PstI-KpnI fragment into the SmaI site of pGEX-3X (Pharmacia) to generate a GST-ASX fusion. The fusion product was expressed in E. coli AD202, affinity purified over glutathione-agarose (Pharmacia) and injected into animals. Immunoglobulins were purified from serum using T-gel (Pierce Biochemical) and used directly for staining embryos and polytene chromosomes. An amino terminal antibody was raised in sheep against the peptide THSLRRHLPRIIVKPIPPEKKG beginning at amino acid 77 in ASX, and affinity purified on a column containing the peptide used as an immunogen and supplied to us by Chiron Mimotopes. An antibody against β-galactosidase was obtained from Jackson Laboratories.

Antibodies raised against both epitopes immunoprecipitated in vitro translated ASX protein. Neither the mouse antibody to the carboxy-terminus, nor the sheep anti-peptide antibody reacted with Asx null embryos, showing that they do not cross-react with other proteins (data not shown). We depleted the rabbit serum by reacting it with the original immunogen and this depleted serum gave no signal on polytene chromosomes, demonstrating that the polytene signals are specific to ASX and are not due to cross-reaction (data not shown). Antibodies were reacted with embryos (McKeon and Brock, 1991) and polytene chromosomes (DeCamillis et al., 1992), as described previously, except that some polytene nuclei were fixed in 1.0% instead of 3.7% paraformaldehyde for 15 seconds prior to squashing. The rabbit serum was used for mapping ASX-binding sites, but similar results were obtained with the sheep antibody.

We reported previously the recovery of the Asx^P1 mutation generated by P element insertion in region 51A5-6, and subsequent loss of the P element in revertants (Sinclair et al., 1992). Comparison of the DNA obtained from Asx^P1 homozygotes and revertant flies by Southern analysis allowed us to identify the insertion site responsible for the Asx mutation. A restriction fragment containing the P element and the flanking DNA was isolated from genomic DNA, and the flanking DNA was used subsequently to screen a genomic cosmid library. Cosmids spanning about 55 kb were recovered. Reverse northern analysis was carried out using labelled cDNA prepared from embryo and adult RNA to identify which fragments were transcribed. Subsequently, these restriction fragments were used to probe northern blots of embryonic poly(A)⁺ RNA.

The P element associated with the Asx^P1 allele lies in the 4.5 kb EcoR1 fragment that hybridizes to two divergent, non-overlapping transcripts, a 7.0 kb transcript that extends distally, and a 4.4 kb transcript that extends proximally (Fig. 1A). We show elsewhere that the proximal transcript encodes the Drosophila homologue of cleavage and polyadenylation specificity factor (Salinas et al., 1998). The P element was inserted within the distal transcript, which seemed likely to correspond to Asx. To confirm this, 10 Asx mutants were analyzed for the presence of genomic rearrangements and for the presence of altered transcripts. Asx³ has an approximately 1.3 kb deletion that maps to the 2.3 kb SalI fragment within the 6.0 kb EcoRI fragment immediately distal to the P element (Fig. 1B). As expected, northern analysis of this mutant shows the presence of a 5.7 kb band in mRNA prepared from Asx³ heterozygotes, which correlates with the deletion predicted from Southern analysis (Fig. 1B). No changes were observed in the other mutants, or in the proximal transcript in Asx³ mutants. Taking the P element insertion data and the mutant analysis together, we conclude that the distal transcript encodes Asx.

An imaginal disc cDNA library was screened for cDNAs corresponding to the distal transcript. Six cDNAs were recovered, providing a total of 6.6 kb. The cDNA was sequenced and compared to the genomic sequence. For most of development, Asx encodes a single transcript that contains 11 introns (Fig. 1A). The 3′-most cDNA does not contain a poly(A) tail, but 100 bp downstream in the genomic sequence is a canonical AATAAA, which likely indicates the 3′ end of the gene. To determine the 5′ end of the transcript, a 5′ RT-PCR reaction was used to extend the available cDNA by 200 bp. The total cDNA of 6.8 kb, added to the genomic DNA including the polyadenylation signal gives a transcript of 6.9 kb, in good agreement with the northern data. The P element is inserted 182 bp downstream of the 5′ end of the cDNA, 256 bp upstream of the translation start. Only 350 bp separate the 5′ ends of the proximal and distal transcripts (Fig. 1A).

The conceptual translation of Asx is presented in Fig. 2. The open reading frame begins 438 bp downstream of the putative transcription start and continues for 1668 amino acids. The protein has an estimated relative molecular mass of 182,000 and an estimated pI of 6.8. At the extreme carboxy terminus is a group of 8 cysteines with the structure CxCx₇CxxCxxxCxxxCx₆CxxC, which could be represented as two consecutive Zn fingers. Another notable feature is the distribution of specific amino acids, exemplified by a stretch of 20 alanines near the amino terminus of the protein and an A₃VA₃ stretch within the carboxy terminal region. There are at least 7 regions with glutamine repeats or a high percentage of glutamines. Glutamine and alanine together make up more than 24% of the amino acids, and proline, serine and threonine contribute 8.1, 7.8 and 6.6% respectively. Serine and threonine can occur in a localized area: 10/21 amino acids within residues 152-173 and 15/30 amino acids within residues 920-950 are S or T. These regions, plus two putative nuclear localization signals, are also marked on the diagram in Fig. 2B. Additional structural features include a nucleotide-binding motif at residues 1033-1039 (Koonin, 1993) and the RGRP tetrapeptide AT hook motif at residues 1502-1505 (Ashley et al., 1989).

Sequences in the mouse and human EST databases conserve the spacing of the cysteines in ASX and also conserve 25/32 amino acids, but this cysteine cluster is not related to any previously characterized cluster of cysteines in the database. A domain of unknown function near the amino terminus of ASX is conserved in sequences found in the mouse and human EST databases. The domain extends from residues 210-330 in ASX, but is somewhat shorter in the mammalian homologues. A comparison of mammalian and ASX conserved domains is shown in Fig. 2C,D. The mouse and human homologues of Asx have been cloned and characterized (F. R., unpublished data). It appears likely that ASX, like other PcG proteins, has a conserved function in flies and mammals.

To determine if Asx is developmentally regulated, we examined its steady-state RNA levels at different developmental stages using northern analysis. As shown in Fig. 3, high levels of Asx transcripts are present in embryos 0-1.5 hours post egg deposition, then levels fall sharply in 1.5-3 hour embryos, before rising through the remainder of embryogenesis. Transcript levels are low in larvae, and increase again in pupae and adults. Three transcripts are detected in adult males, which have not been further characterized, but which may indicate alternate splicing.

We also examined the spatial regulation of Asx transcripts. As shown in Fig. 4A, Asx RNA is abundant in nurse cells, but appears to be absent from stage 10 oocytes. Because newly laid embryos have high Asx RNA levels, it appears likely that Asx mRNA is deposited in oocytes late in oogenesis. At the blastoderm stage, Asx mRNA is more abundant in a broad band in the anterior region of the embryo, and in a narrower band in the posterior region (Fig. 4B). This pattern is similar to that seen with polyhomeotic mRNA in blastoderm embryos (Deatrick, 1992; DeCamillis and Brock, 1994). During the remainder of embryogenesis, Asx RNA is ubiquitous, although it rapidly becomes much more highly expressed in the neurectoderm, and later in the CNS (Fig. 4C).

We used the mouse antibody described in the Materials and Methods to investigate ASX distribution in embryos. We did not detect a high concentration of ASX in nuclei prior to the cellular blastoderm stage (Fig. 4D) but, by the cellular blastoderm stage, ASX is detectable both in nuclei and cytopolasm. At this stage, there is a marginal increase in ASX in the anterior portion of the embryo, but this spatial regulation is not as easily seen for the ASX protein as for the mRNA. Later in embryogenesis, ASX is ubiquitous, but not uniformly distributed; protein is more heavily concentrated in the neurectoderm and the CNS than in other tissues (Fig. 4E,F), whereas staining levels are low in the amnioserosa.

We have noted previously that homozygous Asx^P1 adults show anterior as well as posterior homeotic transformations (Sinclair et al., 1992). Therefore it was of interest to determine if we could detect differences in the regulation of Ubx in Asx^P1 embryos relative to embryos homozygous for a deficiency uncovering Asx (Df(2R)trix) or for a gain-of-function Asx (Asx³) mutation. We monitored Ubx expression using the bxd14 element, which carries a Polycomb Group Response Element (PRE) and is included in a 14.5 kb fragment from the bxd regulatory region of Ubx that regulates a lacZ reporter in parasegment 6-13 (Simon et al., 1990). This element has been used by Soto et al. (1995) to show that Asx is active in the CNS. As shown in Fig. 5, compared to wild-type, both Asx^P1 and Df(2R)trix embryos show derepression of the lacZ reporter in the central nervous system in parasegments 2-5. Similar results were obtained with Asx³ (results not shown). We see no evidence of reduced expression of the bxd14 reporter in parasegments 6-13. We conclude that Asx^P1 mutants, like other Asx mutants, cause derepression of the bxd14 reporter in the anterior CNS during embryogenesis.

Soto et al. (1995) have reported that in Asx mutants derived from homozygous mutant germlines, there is ectopic expression of endogenous ABD-B in the epidermis and visceral mesoderm, but not in the central nervous system. As noted above, they and we were able to detect a requirement for Asx in the central nervous system using the bxd14 element that contains a Ubx regulatory element. Therefore, we decided to re-examine the requirement for Asx in the CNS for regulation of endogenous Ubx. Fig. 5 shows that Ubx is ectopically expressed in parasegment 3 in stage 15 embryos in two Asx mutants, and thus that Asx is required for regulation of the endogenous Ubx gene in the CNS. However, Asx^P1Ubx gene (data not shown).

All PcG proteins tested so far are chromatin proteins that bind to discrete euchromatic sites, many of which overlap among different PcG members. Asx has unique characteristics that indicate it may be functionally distinct from other PcG proteins, so polytene chromosomes were stained with the antibodies described above to determine if the ASX polytene staining pattern differed from that of other PcG proteins.

We reliably detected 90 sites of antibody staining on polytene chromosomes (Fig. 6A). Table 1 lists the cytological locations of the ASX-binding sites and compares them to the binding sites of PH/PC/PCL. Of the 90 sites, 27 are unique to ASX and 63 overlap with PC/PH-binding sites. PC and PH bind an additional 38 sites not recognized by ASX. A comparison of ASX and PH-binding sites is shown in Fig. 6 for part of the X chromosome. In addition to the differences in binding sites, there are also differences in staining intensity at specific sites. Sites 48A, 49EF and 100A all stain very intensely with antibodies to PH or PC but stain very weakly with ASX, whereas sites 35AB, 56C and 93E stain very intensely for ASX but weakly for PH or PC. Only two of the unique ASX sites overlapped sites unique to PSC, which also demonstrates about 70% overlap with PH/PC (Rastelli et al., 1993). One of the common ASX-binding sites is 51A, which is the location of the Asx locus, suggesting that ASX, like other PcG genes (DeCamillis et al., 1992), may be autoregulated.

The ASX protein shares structural features with other members of the PcG with respect to glutamine and serine/threonine repeats as in PH (DeCamillis et al., 1992), and a cysteine cluster like many PcG proteins including PSC, PCL, PH and SCM (Simon, 1995). The 32 amino acid cysteine cluster at the carboxy terminus of ASX is conserved at all 8 cysteine residues, and 25/32 residues overall relative to a mammalian EST sequence, suggesting that this domain has a conserved function. It is likely that the cysteine clusters are protein interaction domains, such as those found in the SV40 large T antigen (Loeber et al., 1991), or the double zinc finger motif human cysteine-rich protein (Feuerstein et al., 1994). One prominent feature of ASX is the presence of 20 consecutive alanines near the amino terminus. Enhancer of Polycomb (E(Pc)) contains a region with 18 alanines in a 21 amino acid stretch (K. Stankunas and H.W.B., unpublished data). In addition, cramped, a newly reported PcG gene contains a 36 aa sequence with 40% alanine (Yamamoto et al., 1997) and Sex combs on midleg contains a 29 aa region that is 52% alanine (Bornemann, 1996). Alanine-rich regions have been implicated in the repression functions of both yeast and Drosophila transcription factors (Tzamarias and Struhl, 1994 and references therein), although long uninterrupted runs of alanine have not been observed in these cases. The function of alanine repeats in PcG proteins remains to be demonstrated.

Flies homozygous for Asx^P1 alleles of Asx exhibit both posterior and anterior transformations, but mapping of the P element insertion site in Asx^P1 mutants has not clarified the mechanism. The P element is inserted into the 5′ untranslated region of the gene and thus it should not interfere with the protein structure of ASX. The presence of the P element does not cause a change in RNA length, suggesting that the element is spliced out. Nor does the P element cause a reduction in the steady state concentration of Asx mRNA (D. A. R. S., unpublished observations). Heterozygous deficiencies for Asx do not exhibit anterior transformations, but do enhance anterior transformation phenotypes of heterozygous mutations in trithorax, as do Asx^P1 mutations (T. A. M., unpublished data).

Asx^P1 homozygotes show anterior transformations only in adults. We confirmed that Asx^P1 homozygotes exhibit ectopic expression of the bxd14 reporter, similar to that seen in other Asx mutations. Despite the strong effect that the Asx^P1 mutation has on the regulation of the bxd14 reporter gene, homozygous Asx^P1 embryos displayed no defect in the regulation of the endogenous Ubx gene in the CNS, since we did not detect any reduction in bxd14 reporter or endogenous Ubx expression in parasegments 6-13. It may be that Asx has a later function in imaginal tissues that is specific for activation of homeotic genes, or that the embryo assay with bxd14 or Ubx expression is not sufficiently sensitive to detect reductions in expression.

The expression patterns of Asx mRNA and protein in ovaries and embryos is very similar to that reported for other PcG genes (Paro and Zink, 1992), (DeCamillis and Brock, 1994), (Martin and Adler, 1993; Lonie et al., 1994). One potential explanation for the observation that Asx regulates Abd-B genes in the epidermis, but not in the central nervous system (Soto et al., 1995), is that the expression of Asx is tissue-limited. Our observations do not support this hypothesis, however, because the expression of Asx, like that of other PcG genes, is ubiquitous.

The results showing tissue-specificity of ASX raise an interesting problem for the current view of how PREs function in the BX-C. There appears to be approximately one strong PRE per parasegmental domain, measured either by direct binding of PC to BX-C DNA (Chiang et al., 1995; Strutt et al., 1997), or by functional assays (Simon et al., 1993). It appears that each PRE can repress the various stage- and tissue-specific enhancers within the parasegment (Chan et al., 1994; Pirrotta et al., 1995), and also repress heterologous enhancers if experimentally manipulated (Simon et al., 1993), (Chiang et al., 1995). This implies that PREs do not provide specificity and that their activity depends on the state of activity when the PcG complex is formed (Poux et al., 1996). Asx mutations have no effect on the regulation of most of the homeotic loci in the CNS, but show derepression phenotypes in the epidermis that are as strong as many other PcG genes. This tissue specificity of Asx mutations implies that Asx may differentially mediate interactions between a PRE and specific enhancers, and thus in some way provide specificity of activity to the PRE. One possibility is that multiple PcG complexes assemble at each PRE, and that different complexes interact with different enhancers.

The Ubx gene is unusual in that it requires wild-type Asx function for proper spatial restriction in the CNS, but the CNS derepression phenotype of Ubx in Asx mutants is relatively mild and restricted to a few individually staining cells in PS3 and PS4. Conversely, the bxd14 reporter shows strong derepression in the CNS in Asx mutants. Mutations in Pc and esc differ from Asx mutations in that they show approximately equal derepression phenotypes for the bxd14 reporter and the endogenous Ubx gene. The fact that CNS expression of the bxd14 reporter is more sensitive to Asx mutations than the endogenous Ubx gene indicates that at least in relation to Asx activity, functional interactions may exist between the bxd14 element and other elements at the Ubx locus. Such interactions may help stabilize PRE activity at the endogenous locus, but when the bxd14 element is isolated these stabilizing interactions are not present and thus it becomes more sensitive to Asx activity.

Our polytene staining results show that ASX binds to the sites of the Antennapedia complex and to the bithorax complex (BX-C), consistent with the homeotic phenotypes, and reports of ectopic expression of homeotic genes in Asx mutants (McKeon and Brock, 1991; Simon et al., 1992; Soto et al., 1995). The apparent exception is that Asx does not appear to be required for regulation of Scr, abd-A or Abd-B in the CNS (Soto et al., 1995). Nevertheless, Asx is required for regulation of endogenous Ubx in the central nervous system, suggesting that it specifically regulates Ubx. This result implies that different PcG proteins can bind to different PREs within the BX-C.

We have shown that Asx mutations strongly enhance mutations in most other PcG genes (Campbell et al., 1995). One explanation for these data is that PcG proteins participate in multimeric complexes (Franke et al., 1992). Consistent with this idea, ASX protein partially co-maps with PH, PC and PCL proteins on polytene chromosomes. The finding that PSC binds to multiple PC/PH sites on polytenes but also has many unique binding sites shows that there are different PcG complexes that bind to different sites. The ASX-binding sites that differ from PC/PH-binding sites and from most PSC-binding sites indicate that ASX can bind to polytene chromosomes in the absence of PC, PH or PSC. This fact could explain some of the unique Asx phenotypes. Because many PC/PH-binding sites do not include ASX, ASX is not required in all cases for PC/PH activity. Conversely, there must be different classes of PH/PC complexes. One complicating factor is that there is no way of knowing if the PcG proteins bound at a given polytene site are members of functional repressive complexes. For example, all PcG proteins tested bind to site 2D, the site of the ph locus and yet ph is obviously not repressed in salivary glands. A similar argument can be made about site 51A, the location of the Asx locus. Nevertheless, our polytene binding data, combined with the previous results of others, argue strongly for heterogeneity of PcG complexes and for the possibility of discrete complexes with different functions. Because ASX binds to many unique sites, we argue that ASX is likely to have different functions than other PcG proteins.

Asx mutations have weaker homeotic phenotypes than mutations in some other PcG genes. Our results, and those of Soto et al. (1995), suggest that one reason is that homeotic genes are differentially sensitive to mutations in Asx within a given tissue; another reason may be that homeotic genes exhibit tissue-specific responses to Asx. One way to acount for these data is to postulate that Asx mediates interactions between specific enhancers and the complexes bound to a parasegmental PRE or PREs. In this view, ASX would not be an obligate member of repressing core PcG complexes containing PH and PC, but could associate with specific PcG complexes binding to a PRE, or with specific enhancers, to mediate interaction between proteins bound to the PRE and to the enhancer, determining whether a given enhancer was or was not repressed by PcG proteins. It will be interesting to determine if ASX protein binds the PRE and if its binding sites overlap with those of other PcG proteins.

Asx^P1 mutations exhibit anterior as well as posterior transformations (Sinclair et al., 1992), and most Asx mutations tested enhance homeotic transformations of trx mutations (T. A. M and H. W. B, unpublished observations). However, Asx^P1 and all other Asx alleles tested also enhance homeotic transformations of other PcG genes (Campbell et al., 1995; Jurgens 1985; T. A. M. and H. W. B., unpublished observations). This indicates that Asx has an important role in both activation and repression of homeotic loci. Interestingly, another PcG gene, Enhancer of zeste also appears to have a dual role in activation and repression of homeotic loci (LaJeunesse and Shearn, 1996). It is possible that ASX (and by extension E(Z)), is a member of both PcG and trithorax group complexes and thus has a dual role in activation and repression. Consistent with the idea that ASX functions as a mediator between protein complexes at the PRE, it is also possible that ASX mediates interactions between PcG and trithorax group proteins to determine if repression by PcG proteins or activation by trithorax group proteins dominates at target sites. ASX could be required for a step that precedes both activation and repression, and thus mutations in Asx could affect both processes. Whatever its function, the fact that ASX distribution on polytene chromosomes does not completely overlap with other PcG proteins argues that ASX has a specific rather than a general role in repression and activation.

The PcG proteins are normally considered to be specific repressors, because PH and PC bind to about 100 sites on polytene chromosomes. Yet if the unique sites of PSC and ASX are considered in addition, the number of PcG targets is increased to about 160, suggesting by extension that many PcG targets remain to be identified. Our data argue strongly against the idea that there is only one way to construct a PcG complex. Instead, it seems likely that a group of related complexes, probably with analogous functions, can be assembled from different components, some of which are shared between complexes, and some of which are not. The shared proteins are likely to constitute a basic structural core and the unique proteins may provide the ability to modulate the activity or specificity of the complex. Because there are ASX-binding sites on polytene chromosomes not recognized by PH, PC, PCL or PSC, we predict that there must be complexes constructed from very different components, that likely have different core proteins. In this model, it may turn out that many genes can be repressed by PcG-like complexes, and that the original definition of PcG genes as repressors of homeotic loci will turn out to be too limited. It will be interesting to see if novel complexes containing a subset of PcG proteins function in repression. For the future, an important goal will be to understand the mechanisms that account for assembly of different PcG complexes at different targets, to determine the role of individual constituents of PcG complexes and to determine if different PcG complexes have different functions.

This work was supported by grants from the National Cancer Institute of Canada to H. W. B. We thank Welcome Bender for stocks.